College laboratory safety management and risk identification intelligent agent platform

Abstract

The invention provides a college laboratory safety management and risk identification intelligent agent platform, which relates to the technical field of safety management, and comprises the following steps: obtaining environmental parameters through edge computing nodes to carry out real-time risk analysis, and setting multi-level safety threshold grading early warning; a mixed decision framework combining a graph neural network and reinforcement learning is adopted, a dynamic reasoning mechanism is constructed based on a security management knowledge graph, and intelligent research and judgment and co-processing of multi-dimensional security risks are realized; and finally, feeding back a disposal result to a safety management personnel terminal. According to the invention, the laboratory safety risk identification precision and the emergency disposal efficiency are improved, and intelligent safety management is realized.

Description

Technical Field

[0001] The present invention relates to the field of safety management technology, and in particular to an intelligent platform for safety management and risk identification in university laboratories. Background Art

[0002] Traditional university laboratory safety management relies primarily on manual inspections and fixed management protocols, which are unable to promptly detect and address various safety risks during experiments. In recent years, with the development of technologies such as the Internet of Things and artificial intelligence, some universities have begun to experiment with intelligent management systems. These systems collect laboratory environmental data through sensor networks and integrate cloud computing platforms for data analysis and risk assessment, aiming to improve the efficiency and accuracy of laboratory safety management.

[0003] However, the existing university laboratory safety management technology still has the following defects and deficiencies: Existing technologies mainly adopt a centralized cloud computing architecture, and all data needs to be uploaded to a remote server for processing, resulting in a high system response delay and the inability to achieve real-time monitoring and rapid response to sudden security incidents. Especially when the network connection is unstable, the efficiency of security risk handling is significantly reduced.

[0004] Most existing laboratory safety management systems use a simple threshold judgment mechanism, lacking the ability to fine-tune the classification and intelligent analysis of safety risks. They are unable to automatically adjust response strategies according to different risk types and severities, resulting in a "one-size-fits-all" safety management measure that is neither economical nor efficient. Summary of the Invention

[0005] The embodiments of the present invention provide an intelligent platform for university laboratory safety management and risk identification, which can solve the problems in the existing technology.

[0006] A first aspect of an embodiment of the present invention provides a university laboratory safety management and risk identification intelligent platform, including: Obtain the environmental parameters of the target laboratory, conduct real-time security risk analysis through edge computing nodes, set multi-level security thresholds within the edge computing nodes, and classify the abnormality levels of different environmental parameters. When the environmental parameters exceed the corresponding level of security thresholds, the edge computing nodes generate risk warning information of the corresponding level and trigger emergency response instructions; The risk warning information is sent to the laboratory's corresponding decision-making execution agent. The decision-making execution agent adopts a hybrid decision-making framework that combines graph neural networks and reinforcement learning. By establishing a laboratory safety management knowledge graph, the relationship between various safety factors in the laboratory is modeled. Based on the safety management knowledge graph, a dynamic reasoning mechanism is constructed to achieve intelligent analysis and coordinated disposal of multi-dimensional safety risks. The coordinated disposal includes equipment linkage control and environmental parameter adjustment. The execution process and effect of the collaborative disposal are recorded, and the disposal results are fed back to the mobile terminal of the laboratory safety manager.

[0007] Obtain the environmental parameters of the target laboratory, conduct real-time security risk analysis through edge computing nodes, set multi-level security thresholds within the edge computing nodes, and classify the abnormality levels of different environmental parameters, including: Transmitting the environmental parameters to an edge computing node, wherein the edge computing node performs standardization processing on the environmental parameters and removes outliers to generate preprocessed environmental parameters; Based on the preprocessed environmental parameters, calculating statistical characteristic values ​​and time series characteristic values ​​of the environmental parameters; constructing a multi-level safety threshold system for the statistical characteristic values ​​and the time series characteristic values, the multi-level safety threshold system including a normal operation threshold, an early warning threshold, a warning threshold, and a danger threshold; The risk level of the environmental parameter is calculated, and the risk level is determined by the degree of deviation of the current value of the environmental parameter from the normal range, the rate of change of the environmental parameter, the duration of the continuous abnormality of the environmental parameter, and the coupling influence of the environmental parameter with other environmental factors.

[0008] The decision-making execution agent adopts a hybrid decision-making framework that combines graph neural networks and reinforcement learning. By establishing a laboratory safety management knowledge graph, it models the relationship between various safety factors in the laboratory and builds a dynamic reasoning mechanism based on the safety management knowledge graph to achieve intelligent analysis and coordinated handling of multi-dimensional safety risks, including: Construct a safety management knowledge graph for the laboratory. The safety management knowledge graph includes equipment nodes, environment nodes, personnel nodes, and security event nodes. The nodes are connected by association edges, which describe the interaction mode and impact degree between the nodes. Based on the security management knowledge graph, real-time node status data is collected, the real-time node status data is input into the graph neural network, and node feature vectors are extracted by calculating the attention coefficient between nodes. The attention coefficient represents the association strength between different nodes, and the node feature vector includes the temporal characteristics and relationship characteristics of the nodes; Identify potential safety risks according to the node feature vector, construct a state space based on the risk identification result, the state space including a current safety state vector and a predicted safety state vector, and generate a set of selectable disposal solutions as an action space; Each disposal plan in the disposal plan set is evaluated, and the evaluation indicators include the degree of risk elimination, disposal timeliness and resource consumption. The evaluation results are used as the basis for calculating the reward function; the disposal plan is optimized using the value function iteration method, and the optimal disposal strategy is calculated based on the current state and the reward function. The optimal disposal strategy takes into account the weight ratio of multiple disposal targets.

[0009] Each disposal solution in the disposal solution set is evaluated, and the evaluation result is used as the basis for calculating the reward function; the disposal solution is optimized using the value function iteration method, and the optimal disposal strategy is calculated according to the current state and the reward function, including: Standardizing the evaluation indicators of the treatment plans, setting weight coefficients for the standardized evaluation indicators, and constructing a reward function through weighted combination, wherein the reward function is used to quantify the comprehensive performance of different treatment plans; Determine the current laboratory state based on the safety monitoring data of the target laboratory, input the current laboratory state and the reward function into a value function iterator, and calculate the state value function based on the state transition probability. The state value function represents the long-term benefits of taking different treatment plans under the current state; Calculating an action-value function using a temporal difference learning method. The action-value function combines the immediate reward and the maximized action value of the future state. The update speed of the value function is controlled by a learning rate parameter. The action-value function quantifies the value of each disposal action. A laboratory safety disposal strategy is generated based on the action-value function, and a temperature parameter is introduced to convert the deterministic strategy into a probability distribution form to achieve diversity in disposal scheme selection. The disposal strategy guides a specific safety risk disposal process.

[0010] The platform is also used to: During the security risk handling process, a state transition sequence and a reward value sequence are recorded, and a strategy performance metric is calculated based on the state transition sequence and the reward value sequence, wherein the strategy performance metric represents the overall effect of the handling strategy; The policy gradient method is used to optimize the policy network parameters, and the optimized policy network parameters are stored in the policy library to guide subsequent laboratory safety risk disposal.

[0011] Intelligent analysis and coordinated handling of multi-dimensional security risks include: Calculating the degree of correlation between the safety monitoring data of the target laboratory, and constructing a risk association relationship based on the degree of correlation, wherein the risk association relationship includes a time dimension association and a space dimension association, and setting weights for the time dimension association and the space dimension association; Classify security risks according to the risk association relationship, set benchmark thresholds for different risk types, and determine the risk level in combination with the risk association relationship when the security monitoring data exceeds the corresponding benchmark threshold; Generate a disposal instruction based on the risk level, the disposal instruction including a risk elimination instruction and a resource allocation instruction, and determine the priority of the disposal instruction according to the laboratory safety management regulations; Executing the processing instructions, recording individual processing results during the processing, and calculating a synergy effect based on the degree of coordination between the individual processing results, wherein the synergy effect reflects the comprehensive performance of the combined execution of multiple processing instructions; The treatment strategy parameters are adjusted according to the synergistic effect, the adjustment amplitude of the parameters is controlled by setting the learning step size, the treatment effect after adjustment is recorded, and the treatment strategy is continuously optimized.

[0012] The beneficial effects of this application are as follows: The present invention conducts real-time safety risk analysis through edge computing nodes, sets multi-level safety thresholds to perform graded processing of abnormal environmental parameters, realizes timely discovery and accurate early warning of laboratory safety risks, and improves the accuracy of risk identification and response speed.

[0013] This invention adopts a hybrid decision-making framework that combines graph neural networks and reinforcement learning to construct a laboratory safety management knowledge graph and dynamic reasoning mechanism, which can realize intelligent analysis and collaborative disposal of multi-dimensional safety risks, thereby improving the intelligence level and disposal efficiency of laboratory safety management.

[0014] By recording the execution process and effects of collaborative disposal and feeding back the disposal results to the mobile terminal of the safety manager, the present invention realizes the traceability and closed-loop management of the entire process of safety risk disposal, effectively promotes the accumulation and optimization of laboratory safety management experience, and improves the overall safety management level. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the university laboratory safety management and risk identification intelligent platform according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0017] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0018] Figure 1 This is a flow chart of the university laboratory safety management and risk identification intelligent platform according to an embodiment of the present invention. Figure 1 As shown: Obtain the environmental parameters of the target laboratory, conduct real-time security risk analysis through edge computing nodes, set multi-level security thresholds within the edge computing nodes, and classify the abnormality levels of different environmental parameters. When the environmental parameters exceed the corresponding level of security thresholds, the edge computing nodes generate risk warning information of the corresponding level and trigger emergency response instructions; The risk warning information is sent to the laboratory's corresponding decision-making execution agent. The decision-making execution agent adopts a hybrid decision-making framework that combines graph neural networks and reinforcement learning. By establishing a laboratory safety management knowledge graph, the relationship between various safety factors in the laboratory is modeled. Based on the safety management knowledge graph, a dynamic reasoning mechanism is constructed to achieve intelligent analysis and coordinated disposal of multi-dimensional safety risks. The coordinated disposal includes equipment linkage control and environmental parameter adjustment. The execution process and effect of the collaborative disposal are recorded, and the disposal results are fed back to the mobile terminal of the laboratory safety manager.

[0019] In an optional implementation, environmental parameters of the target laboratory are obtained, real-time security risk analysis is performed through edge computing nodes, multi-level security thresholds are set within the edge computing nodes, and abnormality levels of different environmental parameters are graded, including: Transmitting the environmental parameters to an edge computing node, wherein the edge computing node performs standardization processing on the environmental parameters and removes outliers to generate preprocessed environmental parameters; Based on the preprocessed environmental parameters, calculating statistical characteristic values ​​and time series characteristic values ​​of the environmental parameters; constructing a multi-level safety threshold system for the statistical characteristic values ​​and the time series characteristic values, the multi-level safety threshold system including a normal operation threshold, an early warning threshold, a warning threshold, and a danger threshold; The risk level of the environmental parameter is calculated, and the risk level is determined by the degree of deviation of the current value of the environmental parameter from the normal range, the rate of change of the environmental parameter, the duration of the continuous abnormality of the environmental parameter, and the coupling influence of the environmental parameter with other environmental factors.

[0020] In this embodiment, a method for real-time security risk analysis using edge computing nodes is provided. This method first obtains laboratory environmental parameters, including temperature, humidity, air pressure, hazardous gas concentrations, and noise levels. These environmental parameters are collected in real time by a sensor network distributed throughout the laboratory and transmitted to the edge computing nodes via a secure data transmission protocol.

[0021] After receiving environmental parameters, the edge computing node first performs normalization and outlier removal. Normalization uses the maximum-minimum normalization method to map each environmental parameter to the [0, 1] range for ease of subsequent processing. For example, if the original value collected by the temperature sensor is 28.5°C and the temperature range is set to 0-50°C, the normalized temperature value is 0.57. Outlier removal uses a moving window median filter. For a data sequence with a window size of 5, if the temperature value suddenly jumps from a normal 25°C to an abnormal 45°C at a certain moment, while the preceding and succeeding data points are both within the 24-26°C range, the 45°C point is identified as an outlier and replaced with the window's median value of 25°C.

[0022] After preprocessing, the edge computing node calculates the statistical and time series characteristics of the environmental parameters. Statistical characteristics include mean, standard deviation, and quantile values. Taking hazardous gas concentrations as an example, after 24 hours of continuous monitoring, the system calculated that the mean of carbon dioxide was 800 ppm, the standard deviation was 120 ppm, the 25th percentile was 720 ppm, and the 75th percentile was 880 ppm. Time series characteristics include rate of change, fluctuation frequency, and periodicity. For example, the hourly rate of change of the temperature parameter is 0.5°C / hour, the daily fluctuation is 4°C, and the difference in mean temperature between weekdays and non-workdays is 2.5°C.

[0023] Based on the calculated characteristic values, the system constructs a multi-level safety threshold system. For temperature parameters, the normal operating threshold is set at 22-26°C, the early warning threshold is 20-22°C or 26-28°C, the warning threshold is 18-20°C or 28-30°C, and the danger threshold is below 18°C ​​or above 30°C. For hazardous gas concentrations, taking carbon dioxide as an example, the normal operating threshold is ≤1000ppm, the early warning threshold is 1000-1500ppm, the warning threshold is 1500-2000ppm, and the danger threshold is >2000ppm. For the temperature change rate, the normal operating threshold is ≤1°C / hour, the early warning threshold is 1-2°C / hour, the warning threshold is 2-3°C / hour, and the danger threshold is >3°C / hour.

[0024] The risk level calculation for environmental parameters considers four factors: the degree of deviation of the parameter's current value from the normal range, the rate of change, the duration of the abnormality, and the coupling effects with other environmental factors. The degree of deviation is quantified by dividing the difference between the current value and the midpoint of the normal range by half the width of the normal range. For example, if the temperature is 28°C, the midpoint of the normal range is 24°C, and the width of the normal range is 4°C, the degree of deviation is (28-24) / (4 / 2) = 2, indicating a deviation of 2 standard units. The rate of change is calculated by taking the mean of the first-order differences of the environmental parameter within a specified time window. For example, if the temperature rises from 24°C to 27°C over the past 30 minutes, the rate of change is 0.1°C / minute. The duration of the abnormality records the cumulative time that the parameter exceeds the normal threshold. For example, if the temperature remains above 26°C for 40 minutes. The coupling effects are determined by analyzing the correlation matrix between environmental parameters. For example, a positive correlation coefficient of 0.75 between rising temperature and rising carbon dioxide concentration indicates a strong coupling relationship between the two.

[0025] The final determination of the risk level is based on a weighted scoring method. The weight of the deviation degree is set at 0.4, the weight of the change rate is set at 0.3, the weight of the duration of the abnormality is set at 0.2, and the weight of the coupling impact is set at 0.1. Each factor is scored based on its severity, ranging from 0 to 10 points. A total score below 3 points indicates a normal state, 3-5 points indicates a warning state, 5-8 points indicates a warning state, and above 8 points indicates a dangerous state. Taking the temperature parameter as an example, when the temperature is 28.5°C, the change rate is 0.15°C / minute, the duration of the abnormality is 50 minutes, and the coupling impact with other parameters is medium (score of 6 points), the deviation degree score is 7 points, the change rate score is 8 points, and the duration of the abnormality score is 5 points. The final weighted total score is 0.4 × 7 + 0.3 × 8 + 0.2 × 5 + 0.1 × 6 = 6.9 points, which is considered a warning state.

[0026] Edge computing nodes automatically execute corresponding security response strategies based on risk levels. In the early warning state, the system records the anomaly and sends a notification to the laboratory administrator. In the warning state, the system activates appropriate environmental conditioning equipment (such as air conditioning and exhaust systems) and issues a warning to all personnel on site. In the dangerous state, the system activates the emergency plan, including the safe shutdown of laboratory equipment, personnel evacuation reminders, and the initiation of emergency response measures.

[0027] This method leverages the low-latency characteristics of edge computing to enable real-time analysis and risk assessment of environmental parameters, ensuring safe laboratory operation. The system completes the entire data collection, processing, and risk assessment process every 10 seconds, with a response delay of no more than 200 milliseconds, meeting the real-time requirements of laboratory safety monitoring. Edge nodes aggregate and upload analysis results every hour to cloud servers for long-term data storage and advanced analysis, forming a complete closed-loop safety management system.

[0028] In an optional implementation, the decision-making agent uses a hybrid decision-making framework that combines graph neural networks and reinforcement learning. By establishing a laboratory safety management knowledge graph, the relationship between various safety factors in the laboratory is modeled. A dynamic reasoning mechanism is constructed based on the safety management knowledge graph to achieve intelligent analysis and coordinated handling of multi-dimensional safety risks, including: Construct a safety management knowledge graph for the laboratory. The safety management knowledge graph includes equipment nodes, environment nodes, personnel nodes, and security event nodes. The nodes are connected by association edges, which describe the interaction mode and impact degree between the nodes. Based on the security management knowledge graph, real-time node status data is collected, the real-time node status data is input into the graph neural network, and node feature vectors are extracted by calculating the attention coefficient between nodes. The attention coefficient represents the association strength between different nodes, and the node feature vector includes the temporal characteristics and relationship characteristics of the nodes; Identify potential safety risks according to the node feature vector, construct a state space based on the risk identification result, the state space including a current safety state vector and a predicted safety state vector, and generate a set of selectable disposal solutions as an action space; Each disposal plan in the disposal plan set is evaluated, and the evaluation indicators include the degree of risk elimination, disposal timeliness and resource consumption. The evaluation results are used as the basis for calculating the reward function; the disposal plan is optimized using the value function iteration method, and the optimal disposal strategy is calculated based on the current state and the reward function. The optimal disposal strategy takes into account the weight ratio of multiple disposal targets.

[0029] This invention discloses an intelligent decision-making and execution system for laboratory safety management based on a combination of graph neural networks and reinforcement learning. The system first constructs a laboratory safety management knowledge graph, which contains four core nodes: equipment nodes, environment nodes, personnel nodes, and safety event nodes. Equipment nodes record basic information about laboratory equipment, such as instrument model, usage status, and maintenance records; environment nodes contain environmental parameters such as temperature, humidity, and hazardous gas concentrations; personnel nodes store laboratory staff identity information, safety training records, and operating permissions; and safety event nodes record historical safety incidents, hidden dangers, and their handling. These nodes are interconnected through association edges. For example, "equipment-user-personnel" represents the operational relationship between a specific person and the equipment, while "environment-impact-equipment" represents the degree to which environmental factors affect equipment performance. Weight values ​​are attached to edges to indicate the strength of the association.

[0030] In practical applications, the knowledge graph construction process consists of two phases: data collection and relationship modeling. The data collection phase uses IoT sensors to collect equipment operating data, such as a centrifuge's speed of 5000 rpm, vibration of 0.5g, and temperature of 38°C. The environmental monitoring system collects laboratory environmental data, such as room temperature of 24°C, humidity of 45%, and formaldehyde concentration of 0.08mg / m³. The personnel management system records staff activity, such as technician A entering the lab area at 2:30 PM and operating equipment X. The relationship modeling phase defines the types of associations between nodes and the weight calculation method based on laboratory safety management standards and expert experience. For example, if the centrifuge's vibration exceeds the threshold of 0.8g, the weight of the "failure risk" edge associated with that equipment increases to 0.75, indicating a potential safety hazard.

[0031] The system uses a distributed sensor network to collect node status data in real time. The data collection frequency is dynamically adjusted based on the safety level, with a sampling interval of 5 seconds for critical equipment and high-risk areas and 30 seconds for general areas. The collected data undergoes preprocessing, including data standardization, outlier detection, and missing value filling, before serving as input to the graph neural network. The graph neural network uses an attention mechanism to calculate the weight of information transfer between nodes. For example, if the temperature in a chemical storage area is detected to have risen to 28°C, the system automatically increases the attention coefficient between that environmental node and the surrounding chemical equipment nodes to 0.85, strengthening monitoring of the relevant equipment status.

[0032] The node feature extraction process is implemented using a three-layer graph convolutional network. Each node's feature vector has a dimension of 128 and contains real-time status information and historical evolution trends. For example, the feature vector of a high-pressure reactor includes the current real-time status of 2.5 MPa and 150°C, as well as time series features such as the pressure change rate and temperature fluctuation range over the past hour. It also incorporates relational features such as the operator qualifications associated with the equipment. These multi-dimensional features are aggregated through the information transfer mechanism of the graph structure to generate a comprehensive feature representation.

[0033] Based on the extracted node feature vectors, the system uses a risk assessment model to identify potential safety risks. This model compares node features with pre-set safety thresholds and a rule base to calculate the risk probability and severity. For example, when the system detects that the feature vector of a laboratory ventilation system node indicates a drop in wind speed of 0.3 m / s (below the safety standard of 0.5 m / s) and an increase in hazardous gas concentrations, the risk assessment model calculates a ventilation system failure probability of 0.83 and a "high risk" safety level. The risk identification results are used to construct a state space consisting of a 160-dimensional current safety state vector and a predicted state vector, reflecting the laboratory's overall safety posture.

[0034] The system also generates a set of selectable response options as an action space. For the aforementioned ventilation system failure risk, the system generates multiple alternatives: activating backup ventilation equipment, reducing laboratory activity, and evacuating non-essential personnel. Each response consists of a series of specific steps. For example, "activating backup ventilation equipment" includes three steps: "checking the status of the backup equipment," "switching the control system to the backup line," and "monitoring the ventilation effect after the switch." The design of the action space takes into account the availability of response resources and the feasibility of the operations, eliminating response options that do not comply with laboratory management standards.

[0035] The evaluation of response plans utilizes a multi-dimensional indicator system, including risk mitigation (a normalized value between 0 and 1), response timeliness (response and execution time in minutes), and resource consumption (a comprehensive measure of equipment, manpower, and material consumption). For example, the "activate backup ventilation equipment" plan was evaluated as follows: risk mitigation of 0.95, response timeliness of 2 minutes, and resource consumption (the cost of using the backup equipment plus the operator's man-hours, converted to 12 resource units). These evaluation indicators were combined to create a reward function value of 85 points (out of a maximum of 100).

[0036] The system uses a deep Q network to implement value function iteration. The network structure contains three hidden layers with 256, 128, and 64 neurons in the hidden layers, respectively. By learning from historical disposal cases, the model can predict the long-term benefits of different disposal plans under the current state. The optimization process takes into account the weight ratios of multiple disposal goals. The default configuration is a risk elimination weight of 0.6, a disposal timeliness weight of 0.3, and a resource consumption weight of 0.1. Administrators can adjust these weights according to actual needs. Ultimately, the system selects the disposal plan with the highest Q value as the optimal strategy, and executes the corresponding control instructions through the laboratory management interface to achieve intelligent disposal of safety risks.

[0037] In an optional embodiment, each treatment plan in the treatment plan set is evaluated, and the evaluation result is used as a basis for calculating the reward function; the treatment plan is optimized using a value function iteration method, and the optimal treatment strategy is calculated based on the current state and the reward function, including: Standardizing the evaluation indicators of the treatment plans, setting weight coefficients for the standardized evaluation indicators, and constructing a reward function through weighted combination, wherein the reward function is used to quantify the comprehensive performance of different treatment plans; Determine the current laboratory state based on the safety monitoring data of the target laboratory, input the current laboratory state and the reward function into a value function iterator, and calculate the state value function based on the state transition probability. The state value function represents the long-term benefits of taking different treatment plans under the current state; Calculating an action-value function using a temporal difference learning method. The action-value function combines the immediate reward and the maximized action value of the future state. The update speed of the value function is controlled by a learning rate parameter. The action-value function quantifies the value of each disposal action. A laboratory safety disposal strategy is generated based on the action-value function, and a temperature parameter is introduced to convert the deterministic strategy into a probability distribution form to achieve diversity in disposal scheme selection. The disposal strategy guides a specific safety risk disposal process.

[0038] This invention provides a method for optimizing laboratory safety risk management solutions. In its implementation, each solution in a set of solutions is first evaluated, and the evaluation results are used as the basis for calculating a reward function. The solution is then optimized using a value function iteration method to obtain the optimal strategy.

[0039] Specifically, during the evaluation indicator standardization phase, the system collects the original evaluation indicators for each treatment plan, such as cost, timeliness, and safety improvement indicators. For example, for a chemical laboratory fire risk treatment, the original evaluation indicators for Treatment Plan A were: cost 100,000 yuan, execution time 2 hours, and a safety improvement score of 85 points; the original evaluation indicators for Treatment Plan B were: cost 150,000 yuan, execution time 1.5 hours, and a safety improvement score of 92 points. The system uses the maximum-minimum normalization method to normalize these indicators, mapping all indicator values ​​to the range of 0-1. For example, after standardization of the cost indicator, the standardized cost of Treatment Plan A is 0, and the standardized cost of Treatment Plan B is 1. After standardization of the timeliness indicator, the standardized timeliness of Treatment Plan A is 1, and the standardized timeliness of Treatment Plan B is 0. After standardization of the safety improvement indicator, the standardized safety improvement of Treatment Plan A is 0, and the standardized safety improvement of Treatment Plan B is 1.

[0040] When assigning weights to standardized evaluation indicators, the system determines the importance of each indicator based on the laboratory type and risk characteristics. For the chemical laboratory example above, the weight of the cost indicator is set to 0.2, the weight of the timeliness indicator is set to 0.3, and the weight of the safety improvement indicator is set to 0.5. The reward function constructed through weighted combination is: the reward value for solution A = 0 × 0.2 + 1 × 0.3 + 0 × 0.5 = 0.3; the reward value for solution B = 1 × 0.2 + 0 × 0.3 + 1 × 0.5 = 0.7. This indicates that solution B has better overall performance than solution A.

[0041] To determine the current laboratory status based on the target laboratory's safety monitoring data, the system collects laboratory parameters such as temperature, humidity, gas concentration, and equipment operating status through a sensor network. For the aforementioned chemical laboratory, the current status parameters are: temperature 42°C, methanol vapor concentration 2.5%, and normal firefighting equipment status. The system compares these parameters with preset thresholds and determines that the laboratory is currently in a "moderate fire risk" state.

[0042] After the current lab state and reward function are input into the value function iterator, the system calculates the state value function based on a pre-established state transition probability model. This model describes the probability distribution of lab state changes after adopting a certain action plan in the current state. For example, if action plan A is adopted in the "medium fire risk" state, there is a 60% probability of transitioning to the "low fire risk" state, a 30% probability of remaining in the "medium fire risk" state, and a 10% probability of escalating to the "high fire risk" state. If action plan B is adopted, there is an 85% probability of transitioning to the "low fire risk" state, a 12% probability of remaining in the "medium fire risk" state, and a 3% probability of escalating to the "high fire risk" state.

[0043] The system iterates and calculates repeatedly, updating the estimated value of each state. Initially, the value functions for all states are set to 0. After the first iteration, the state value for the "moderate fire risk" state with solution A is 0.3, and the state value for solution B is 0.7. After multiple iterations, the state values ​​stabilize, ultimately reaching 0.68 for solution A and 0.89 for solution B.

[0044] When calculating the action-value function using the temporal difference learning method, the system combines the immediate reward and the value of the future state to make an estimate. In practice, the learning rate parameter is set to 0.2 and the discount factor is set to 0.9. For solution A under the "moderate fire risk" state, its action-value function is calculated as follows: the immediate reward of 0.3 plus the discounted value of the future state (0.9 multiplied by 0.68), resulting in a current action value of 0.912. For solution B, its action-value function is calculated as follows: the immediate reward of 0.7 plus the discounted value of the future state (0.9 multiplied by 0.89), resulting in a current action value of 1.501.

[0045] When generating a laboratory safety strategy based on the action-value function, the system introduces a temperature parameter, T = 0.5, to transform the deterministic strategy into a probability distribution. For the example above, the probability of selecting option A is 25.7%, and the probability of selecting option B is 74.3%. This probability distribution strategy allows for diverse solution options and helps prevent the system from falling into a local optimum.

[0046] During the execution of the disposal strategy, the system updates laboratory status information in real time and adjusts the disposal strategy based on the new status information. For example, if the temperature is detected to have dropped to 35°C and the methanol vapor concentration has dropped to 1.2%, the system reassesses the risk status to "low fire risk" and generates a new disposal strategy based on the action value function in this state, thus achieving dynamic optimization of the disposal process.

[0047] Through the implementation of the above technical means, the present invention can realize the intelligent optimization of laboratory safety risk management solutions, improve the efficiency and effectiveness of risk management, and reduce the incidence rate and degree of losses of safety accidents.

[0048] In an optional embodiment, the platform is further used to: During the security risk handling process, a state transition sequence and a reward value sequence are recorded, and a strategy performance metric is calculated based on the state transition sequence and the reward value sequence, wherein the strategy performance metric represents the overall effect of the handling strategy; The policy gradient method is used to optimize the policy network parameters, and the optimized policy network parameters are stored in the policy library to guide subsequent laboratory safety risk disposal.

[0049] During the safety risk disposal process, this method comprehensively evaluates and optimizes the effect of the disposal strategy by recording the state transition sequence and reward value sequence, so as to improve the effectiveness of laboratory safety risk disposal.

[0050] During the laboratory safety risk management process, the system continuously records the state transition sequence and the corresponding reward value sequence. The state transition sequence contains state information at each time step during the management process, such as the type, concentration, distribution range, and personnel location of the hazard source within the laboratory. For example, in a laboratory chemical leak scenario, the state might include: the initial chemical leak volume is 500 ml, the diffusion area is 2 square meters, and there are 3 laboratory personnel within 5 meters of the leak point; after the first step of the management action is taken, the chemical leak is controlled, the diffusion area is maintained at 2 square meters, and the personnel have been evacuated to a safe area. This state information is recorded in chronological order, forming a complete state transition sequence.

[0051] The reward sequence records the immediate rewards the system receives after each state transition, reflecting the immediate effectiveness of the response. In the aforementioned chemical leak scenario, the rewards might include: +10 points for successfully controlling the spread of the leak, +15 points for safe evacuation of personnel, and +8 points for correct neutralization measures. These rewards constitute a complete reward sequence.

[0052] Based on the recorded sequence of state transitions and reward values, the system calculates a strategy performance metric. This metric is calculated using a cumulative reward approach. Specifically, all rewards obtained from the initial state to the final state of a complete security risk management process are weighted and accumulated. During this weighting process, a decay factor can be used to discount future rewards, giving near-term rewards a higher weight than far-term rewards. For example, using a decay factor of 0.95, the reward weight obtained in the first step is 1, in the second step it is 0.95, in the third step it is 0.95 × 0.95 = 0.9025, and so on.

[0053] For a complete processing process, assuming there are 5 time steps, the corresponding reward value sequence is [5, 8, 12, 7, 10], and a decay factor of 0.95 is used, the calculated policy performance metric value is: 5×1 + 8×0.95 + 12×0.95 2 + 7×0.95 3 + 10×0.95 4 = 5 + 7.6 + 10.83 + 6.01 + 8.15 = 37.59. This value represents the overall effectiveness of the current policy in the security risk management process. The system can calculate the average policy performance metric through multiple experiments or simulations to more comprehensively evaluate the effectiveness of the management policy.

[0054] To continuously optimize the response strategy, this method uses a policy gradient method to optimize the policy network parameters. The policy network is a parameterized model that takes the current state as input and outputs a probability distribution over possible response actions. The policy gradient method calculates the gradient of the policy performance metric with respect to the policy network parameters and adjusts the parameters along the gradient to maximize the policy performance metric.

[0055] In practice, the policy network can be a three-layer neural network. The input layer corresponds to state features, such as hazard information, environmental parameters, and personnel distribution. The hidden layer contains 128 neurons and uses the ReLU activation function. The output layer corresponds to the probabilities of various action plans and uses the Softmax function to ensure that the output is a valid probability distribution. The parameters of the policy network include the connection weights and bias terms between each layer.

[0056] In the policy gradient method, the system first runs multiple safety risk response simulations or actual operations based on the current policy network parameters, collecting state transition sequences and reward value sequences. For example, for a chemical leak scenario, the current policy network is used to run 10 simulations, recording the complete state transition sequence and reward value sequence for each simulation.

[0057] For each action, the system calculates a policy performance metric for that action and records the state and corresponding action taken at each time step. Based on this information, the system calculates the policy gradient, which is the gradient of the policy performance metric with respect to the policy network parameters.

[0058] For example, a chemical spill response involved five time steps, and the policy performance metric was calculated to be 37.59. The system analyzes the relationship between the probability of selecting a specific action in the current state and the policy performance metric obtained during each time step, and calculates the policy gradient. This gradient reflects how to adjust the policy network parameters to increase the probability of selecting actions that receive high rewards and decrease the probability of selecting actions that receive low rewards.

[0059] After calculating the policy gradient, the system updates the policy network parameters using gradient ascent with a learning rate of 0.01. For example, if the original value of a weight parameter is 0.45 and the corresponding gradient is 0.08, the updated parameter value is 0.45 + 0.01 × 0.08 = 0.4508. The system performs similar updates on all parameters in the policy network.

[0060] After updating the parameters, the system verifies the performance of the new policy. Under the same security risk scenario, the updated policy network is used to perform multiple simulations and calculate the average policy performance metric. If the average policy performance metric improves, the parameter update is effective; otherwise, the learning rate may need to be adjusted or the policy gradient recalculated.

[0061] After multiple rounds of optimization, the system stores the final optimized policy network parameters in the policy library. The policy library can be designed as a parameter database that contains the optimal policy network parameters corresponding to different types of safety risks. Each parameter set includes the weight matrix and bias vector of each layer of the network, stored in binary format. For example, for chemical leakage risk, the policy library stores the corresponding three-layer neural network parameters, including the weight matrix from the input layer to the hidden layer (dimension is the number of input features × 128), the hidden layer bias vector (dimension is 128), the hidden layer to the output layer weight matrix (dimension is 128 × the number of disposal actions) and the output layer bias vector (dimension is the number of disposal actions).

[0062] In subsequent practical applications, when a laboratory safety risk is detected, the system retrieves the policy network parameters corresponding to the risk type from the policy library, rebuilds the policy network model, inputs the current status in real time, obtains the probability distribution of the disposal action, and guides the laboratory safety risk disposal. This approach enables the continuous optimization of disposal strategies and the accumulation of experience, effectively improving the efficiency and success rate of laboratory safety risk disposal.

[0063] In an optional implementation, intelligent assessment and coordinated handling of multi-dimensional security risks include: Calculating the degree of correlation between the safety monitoring data of the target laboratory, and constructing a risk association relationship based on the degree of correlation, wherein the risk association relationship includes a time dimension association and a space dimension association, and setting weights for the time dimension association and the space dimension association; Classify security risks according to the risk association relationship, set benchmark thresholds for different risk types, and determine the risk level in combination with the risk association relationship when the security monitoring data exceeds the corresponding benchmark threshold; Generate a disposal instruction based on the risk level, the disposal instruction including a risk elimination instruction and a resource allocation instruction, and determine the priority of the disposal instruction according to the laboratory safety management regulations; Executing the processing instructions, recording individual processing results during the processing, and calculating a synergy effect based on the degree of coordination between the individual processing results, wherein the synergy effect reflects the comprehensive performance of the combined execution of multiple processing instructions; The treatment strategy parameters are adjusted according to the synergistic effect, the adjustment amplitude of the parameters is controlled by setting the learning step size, the treatment effect after adjustment is recorded, and the treatment strategy is continuously optimized.

[0064] In this embodiment, a method for intelligently assessing and collaboratively addressing multi-dimensional security risks is provided. This method analyzes the correlations between target laboratory safety monitoring data, constructs risk association relationships, and classifies and addresses security risks based on these relationships.

[0065] To calculate the correlation between target laboratory safety monitoring data, the system first collects data recorded by various sensors within the laboratory, including parameters such as temperature, humidity, gas concentration, and pressure. For the temporal dimension, the system analyzes the changing trends of data at the same monitoring point over different time periods. For example, the changing trend of formaldehyde concentration in a chemical laboratory over the past 24 hours is consistent with the temperature change. If the two show a positive correlation, with a correlation coefficient of 0.85 or above, they are considered to have a high temporal correlation. For the spatial dimension, the system compares the distribution of data from different monitoring points over the same time period. For example, the relationship between the ammonia concentration in laboratory area A and the operating status of the ventilation system in area B. If the ammonia concentration in area A increases but the ventilation system in area B fails to increase accordingly, and the correlation coefficient is less than 0.3, the spatial correlation is determined to be weak. After constructing a complete risk correlation, the system assigns a weight of 0.6 to the temporal dimension and a weight of 0.4 to the spatial dimension, based on historical accident data and expert experience.

[0066] When grading safety risks based on the established risk associations, the system sets baseline thresholds for different risk types. For example, for flammable gas leak risk, the alert threshold is set when the monitored gas concentration reaches 20% of the lower explosive limit; for high-temperature risk, an alert is triggered when the temperature exceeds the normal operating temperature of the equipment by 10°C. When safety monitoring data exceeds the corresponding baseline threshold, the system determines the risk level based on the risk associations. For example, if a laboratory detects a methane concentration reaching 25% of the lower explosive limit and the temperature is 12°C higher than normal, given a correlation coefficient of 0.78 between these two parameters, a time dimension weight of 0.6, and a spatial dimension weight of 0.4, the system calculates a comprehensive risk index of 0.82, assigning a "high risk" level. Risk levels are categorized into four levels: low (0-0.3), medium (0.3-0.6), high (0.6-0.85), and severe (0.85-1.0).

[0067] Based on the determined risk level, the system generates corresponding disposal instructions. Disposal instructions include risk elimination instructions and resource allocation instructions. Risk elimination instructions target specific sources of danger, such as "start the exhaust system in area A" and "close gas valve No. 3." Resource allocation instructions involve the allocation of emergency resources, such as "dispatch emergency personnel to area B" and "prepare fire-fighting equipment." The system determines the priority of disposal instructions based on laboratory safety management regulations. For the above-mentioned abnormal methane concentration and temperature, the system first generates an instruction to "start the emergency exhaust system" (priority 5, the highest of 5), followed by "shut down the gas supply" (priority 4), and then "notify the safety management personnel to the scene" (priority 3) and "prepare fire-fighting equipment" (priority 2).

[0068] When executing a response instruction, the system records the execution and effectiveness of each response measure. For example, if the methane concentration drops below 10% of the lower explosive limit within three minutes after the emergency exhaust system is activated, the response is deemed "effective"; if the methane concentration stops rising after the gas supply is shut off, the response is deemed "effective"; and if safety management personnel are notified and no other safety hazards are confirmed on-site, the response is deemed "effective." The system calculates synergy based on the degree of coordination between these individual response results. In this example, the coordination between the exhaust system and the gas supply shutdown is 0.9 (very high), and the coordination with the on-site personnel is 0.7 (high), resulting in a comprehensive synergy score of 0.85, reflecting the overall good performance of the combined execution of multiple response instructions.

[0069] Based on the synergistic effect, the system adjusts the treatment strategy parameters. By setting the learning step size (0.05 in this case), the parameter adjustment range is controlled to avoid drastic fluctuations in the strategy. For the above case, the system maintains the priority confirmation of "activating the emergency exhaust system" at 5, but adjusts the trigger threshold for "shutting down the gas supply" from 20% of the lower explosive limit of methane concentration to 18%, allowing the system to take preventive measures even when the dangerous concentration is low. The system records the effect of the adjusted treatment. The next time a similar situation occurs, the gas supply is shut down in advance, reducing the maximum methane concentration to 21% of the lower explosive limit, thereby reducing the risk level. In this way, the system continuously optimizes the treatment strategy, making risk management more efficient and accurate.

[0070] This method intelligently analyzes the correlation between safety monitoring data, realizes accurate assessment and coordinated disposal of multi-dimensional safety risks in the laboratory, and improves the automation and intelligence level of laboratory safety management.

[0071] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The intelligent platform for university laboratory safety management and risk identification is characterized by: include: Obtain the environmental parameters of the target laboratory, conduct real-time security risk analysis through edge computing nodes, set multi-level security thresholds within the edge computing nodes, and classify the abnormality levels of different environmental parameters. When the environmental parameters exceed the corresponding level of security thresholds, the edge computing nodes generate risk warning information of the corresponding level and trigger emergency response instructions; The risk warning information is sent to the laboratory's corresponding decision-making execution agent. The decision-making execution agent adopts a hybrid decision-making framework that combines graph neural networks and reinforcement learning. By establishing a laboratory safety management knowledge graph, the relationship between various safety factors in the laboratory is modeled. Based on the safety management knowledge graph, a dynamic reasoning mechanism is constructed to achieve intelligent analysis and coordinated disposal of multi-dimensional safety risks. The coordinated disposal includes equipment linkage control and environmental parameter adjustment. The execution process and effect of the collaborative disposal are recorded, and the disposal results are fed back to the mobile terminal of the laboratory safety manager.

2. The platform according to claim 1, characterized in that Obtain the environmental parameters of the target laboratory, conduct real-time security risk analysis through edge computing nodes, set multi-level security thresholds within the edge computing nodes, and classify the abnormality levels of different environmental parameters, including: The environmental parameters are transmitted to an edge computing node, and the edge computing node performs standardization processing on the environmental parameters and removes outliers to generate pre-processed environmental parameters; Based on the preprocessed environmental parameters, calculating statistical characteristic values ​​and time series characteristic values ​​of the environmental parameters; constructing a multi-level safety threshold system for the statistical characteristic values ​​and the time series characteristic values, the multi-level safety threshold system including a normal operation threshold, an early warning threshold, a warning threshold, and a danger threshold; The risk level of the environmental parameter is calculated, and the risk level is determined by the degree of deviation of the current value of the environmental parameter from the normal range, the rate of change of the environmental parameter, the duration of the continuous abnormality of the environmental parameter, and the coupling influence of the environmental parameter with other environmental factors.

3. The platform according to claim 1, characterized in that The decision-making execution agent adopts a hybrid decision-making framework that combines graph neural networks and reinforcement learning. By establishing a laboratory safety management knowledge graph, it models the relationship between various safety factors in the laboratory and builds a dynamic reasoning mechanism based on the safety management knowledge graph to achieve intelligent analysis and coordinated handling of multi-dimensional safety risks, including: Construct a safety management knowledge graph for the laboratory. The safety management knowledge graph includes equipment nodes, environment nodes, personnel nodes, and security event nodes. The nodes are connected by association edges, which describe the interaction mode and impact degree between the nodes. Based on the security management knowledge graph, real-time node status data is collected, the real-time node status data is input into the graph neural network, and node feature vectors are extracted by calculating the attention coefficient between nodes. The attention coefficient represents the association strength between different nodes, and the node feature vector includes the temporal characteristics and relationship characteristics of the nodes; Identify potential safety risks according to the node feature vector, construct a state space based on the risk identification result, the state space including a current safety state vector and a predicted safety state vector, and generate a set of selectable disposal solutions as an action space; Each disposal plan in the disposal plan set is evaluated, and the evaluation indicators include the degree of risk elimination, disposal timeliness and resource consumption. The evaluation results are used as the basis for calculating the reward function; the disposal plan is optimized using the value function iteration method, and the optimal disposal strategy is calculated based on the current state and the reward function. The optimal disposal strategy takes into account the weight ratio of multiple disposal targets.

4. The platform according to claim 3, characterized in that Evaluate each treatment plan in the treatment plan set, and use the evaluation result as a basis for calculating the reward function; The disposal plan is optimized using the value function iteration method. The optimal disposal strategy is calculated based on the current state and reward function, including: Standardizing the evaluation indicators of the treatment plans, setting weight coefficients for the standardized evaluation indicators, and constructing a reward function through weighted combination, wherein the reward function is used to quantify the comprehensive performance of different treatment plans; Determine the current laboratory state based on the safety monitoring data of the target laboratory, input the current laboratory state and the reward function into a value function iterator, and calculate the state value function based on the state transition probability. The state value function represents the long-term benefits of taking different treatment plans under the current state; Calculating an action-value function using a temporal difference learning method. The action-value function combines the immediate reward and the maximized action value of the future state. The update speed of the value function is controlled by a learning rate parameter. The action-value function quantifies the value of each disposal action. A laboratory safety disposal strategy is generated based on the action-value function, and a temperature parameter is introduced to convert the deterministic strategy into a probability distribution form to achieve diversity in disposal scheme selection. The disposal strategy guides a specific safety risk disposal process.

5. The platform according to claim 4, characterized in that The platform is also used to: During the security risk handling process, a state transition sequence and a reward value sequence are recorded, and a strategy performance metric is calculated based on the state transition sequence and the reward value sequence, wherein the strategy performance metric represents the overall effect of the handling strategy; The policy gradient method is used to optimize the policy network parameters, and the optimized policy network parameters are stored in the policy library to guide subsequent laboratory safety risk disposal.

6. The platform according to claim 1, characterized in that Intelligent analysis and coordinated handling of multi-dimensional security risks include: Calculating the degree of correlation between the safety monitoring data of the target laboratory, and constructing a risk association relationship based on the degree of correlation, wherein the risk association relationship includes a time dimension association and a space dimension association, and setting weights for the time dimension association and the space dimension association; Classify security risks according to the risk association relationship, set benchmark thresholds for different risk types, and determine the risk level in combination with the risk association relationship when the security monitoring data exceeds the corresponding benchmark threshold; Generate a disposal instruction based on the risk level, the disposal instruction including a risk elimination instruction and a resource allocation instruction, and determine the priority of the disposal instruction according to the laboratory safety management regulations; Executing the processing instructions, recording individual processing results during the processing, and calculating a synergy effect based on the degree of coordination between the individual processing results, wherein the synergy effect reflects the comprehensive performance of the combined execution of multiple processing instructions; The treatment strategy parameters are adjusted according to the synergistic effect, the adjustment amplitude of the parameters is controlled by setting the learning step size, the treatment effect after adjustment is recorded, and the treatment strategy is continuously optimized.

Images